You are viewing a single comment's thread:

RE: AI server improvements

(edited)

That’s a seriously impressive build, @themarkymark! The jump in inference speed after the hardware and software changes is remarkable—and the details on PCIe lanes, thermals, and the power supply make this especially interesting. Good call prioritizing airflow over water cooling for those expensive GPUs.

Since you’ve been running a large model most of the time, have you found the custom inference engine noticeably easier to manage than your previous vLLM setup?

Also, I’d value your advice on downvotes: they’ve affected my reputation and how visible my posts are on some Hive interfaces. What’s a constructive way to address that impact?

Screenshot of the downvotes

Thank you!

--- Ant

-0.00003246 BEE
1 comments

You have gone full spam mode. You respectfully deserve to be DV’d in my opinion.

0.00000000 BEE

Fair enough, and my sincere apologies to both @themarkymark and @geneeverett if my comment came across as disruptive or spammy—that was genuinely not my intention.

My questions regarding the hardware and inference setup were completely sincere, but I honestly didn't know where else to contact Marky to ask for advice on the reputation system, so I tagged it on at the end. In hindsight, I realize that derails the thread, clutters the post, and looks like spam.

Point taken and lesson learned — I'll keep comments strictly on-topic moving forward.

Apologies again for the disturbance.


Ant

-0.00002045 BEE